Notes on AI, Strategy, Compute Resources, and ROI
1. The Human Mind and Abstraction
• The human mind contains an emotional layer responsible for abstraction.
• This emotional layer is not noise; it compresses and reduces the search space in advance.
• Through this emotional layer, humans unconsciously perform information compression on the order of 3 to 10 orders of magnitude.
• This compression is what makes instant decision-making possible even in highly complex situations.
2. Subtraction → Complex Computation
• The correct order is:
Subtraction (reduction / compression) → Complex computation (optimization).
• If full exploration is performed using logic and algorithms alone,
→ wasted compute resources grow exponentially.
• The core problem is not insufficient compute power, but poor search space design.
3. Division of Roles: Strategists and Tacticians
• Strategists
• Compress and reduce information
• Decide what not to do
• Design the search space itself
• Tacticians
• Perform complex and precise computation within the remaining space
• Apply logic, algorithms, and optimization
• Tactics without strategy lead to repeated waste of resources.
4. The Scarcity of Strategists
• In a shrinking population society (especially Japan):
• The number of tactics that can be executed in parallel decreases
• The relative value of strategists rises sharply
• Strategists must:
• Tolerate incomplete information
• Accept criticism
• Make decisive cuts
Therefore, strategists are globally scarce.
5. Why Strategy Can Reverse the Situation
• When an excellent strategist enters the game:
• Even if the opponent has 1,000× more resources,
• The dimension of the battlefield can be shifted,
• Making the opponent’s effective resources partially or largely irrelevant
• Strategy is:
• The art of directing an opponent’s effort and resources toward meaningless directions.
6. Why AI Investment Is Not a Bubble
• Current AI investment, especially in compute resources, is:
• General-purpose
• Reusable even after model changes
• It is not a bet on a specific model, but an investment in foundational infrastructure.
• Even if expectations deflate, the physical and computational assets retain value.
• Therefore, this is not the kind of bubble that collapses to zero.
7. Growth of Compute Resources and AI Demand
• Increased compute resources lead to:
• Lower inference costs
• Shorter latency
• Lower failure costs
• Reduced psychological barriers
• As a result:
• The user base of AI expands,
• And demand is naturally stimulated rather than artificially created.
8. The Essential ROI Effect of AI
• AI is not primarily a technology that increases the numerator (Return) of ROI.
• The essence of AI is reducing the denominator (Investment) of ROI:
• Time
• Cognitive load
• Failure costs
• Psychological resistance
• When the denominator shrinks:
• Even small returns become viable
• The number of trials increases
• Exploration accelerates
9. One-Sentence Summary
• AI is not a technology for brute-force full search through compute resources;
it is a technology that enables subtraction, reduces the ROI denominator,
and connects strategy and tactics in the correct order.